Author: Mark Young
Date: 18:18:21 10/04/98
Go up one level in this thread
On October 04, 1998 at 13:02:17, Robert Hyatt wrote: >On October 03, 1998 at 09:59:52, Moritz Berger wrote: > >>On October 03, 1998 at 09:54:53, blass uri wrote: >> >>>I did not see a great success for fritz5 in tournaments of computer programs >>>for example you can see the results in Shep's tournament >>>Fritz5 has 1.5 out of 4 >> >>Fritz won a big tournament played by Torsten Schoop, AFAIR. > > >I think *everyone* is looking at the wrong aspect of learning. It is not >about finding "good" openings against a single opponent... it is *much* >more important to detect *bad* openings against *all* opponents. This is >the valuable part of "crafty's book learning" results... to take a huge >collection of PGN games, about 2/3 million games, in fact, and then play from >these games and learn which moves are to be avoided. If you look at my >LearnFunction() you will notice that I don't pay much attention to "winning" >and concentrate on "losing" lines... > >If I can eliminate the losing lines, I identify the "winners" by default. But >mainly I want to eliminate the "losers". It is working, and *anyone* can down- >load my book source, then download my "learning" data and have a book that is >based on 50,000 OTB games played from an opening book of 2/3 million games... > >And not see too many outright blunders as a result... I agree learning would work for both tournament (many different players) and match play (single player). I find this to be true when I play online by hand, so crafty that plays 1000x more and this still holds true I think is pretty conclusive that learning is a valuable part of a programs overall strength. Therefore, leaning should not be circumvented or disabled in anyway in SSDF testing. If other programs do not have learning or have non-effective leaning then those programs need to improve in that area so they can compete effectively in SSDF testing. I do not see the sense in SSDF should try to make a so-called even playing field by handicapping those programs that have added good leaning features to their programs.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.